Skip to main content
Internet Archive's 25th Anniversary Logo

Full text of "Turning to the Dark Side: Evidence for Circumburst Extinction of Gamma-Ray Bursts with Dark Optical Afterglows"

See other formats


o 
o 



o 



Dh' 



Submitted to The Astrophysical Journal 

Turning to the Dark Side: Evidence for Circumburst Extinction of 
Gamma-Ray Bursts with Dark Optical Afterglows 

Daniel E. Reichart^'^ and Sarah A. Yost^ 
ABSTRACT 

We present evidence that the majority of rapidly-, well-localized gamma-ray bursts 
with undetected, or dark, optical afterglows, or 'dark bursts' for short, are most 
likely the result of extinction by dust in the circumburst medium. First, we show 



OO 

CN I that the dark bursts cannot be explained by a failure to image deeply enough quickly 

enough: We fit a variety of brightness distribution models to the optical data, and 

^ ■ find that w 57^^^^;^% of all bursts, and ^ 82^]^-j7% of dark bursts, have afterglows that 

2? • are fainter than R = 24 mag 18 hours after the burst. Secondly, we show that dark 

l/S ■ bursts tend to be X-ray and radio faint. Thirdly, we show that these correlations, 



and more specifically the optical vs. X-ray and optical vs. radio distributions, can 
be explained within the framework of the relativistic fireball model if the dark bursts 
are extinguished by circumburst dust, and the density of the circumburst medium 
i-^ ' spans many orders of magnitude, from densities that are typical of the Galactic disk 

to densities that are typical of dense clouds. Finally, we show that Galactic extinction. 



^ • host galaxy extinction unrelated to the circumburst medium, and the following high 

c/3 . redshift effects - Lyman limit absorption in the source frame, absorption by the 

Lya forest, absorption by excited molecular hydrogen in the circumburst medium, 
and source-frame extinction by the FUV component of the extinction curve - might 
rS ' contribute to the number of dark bursts, but only in small numbers. 

Subject headings: dust, extinction — galaxies: high redshift — gamma rays: bursts 
ISM: clouds — radio continuum: general — X-rays: general 



1. Introduction 

Optical afterglows have been detected for about 1/3 of the rapidly-, well-localized gamma-ray 
bursts (e.g., Fynbo et al. 2001; Lazzati, Covino & Ghisellini 2001). This data-rich subsample of 



^Department of Astronomy, California Institute of Technology, Mail Code 105-24, 1201 East California Boulevard, 
Pasadena, CA 91125 

^Hubble Fellow 



the rapidly-, well-localized bursts has naturally been the focus of the vast majority of the field's 
attention and resources over the past four years, but in the end, it is a biased sample. The nature 
of the so-called 'dark bursts'^ has only recently become a subject of greater interest, and as might 
be expected given that these are by definition data-poor events, contradictory initial findings: 
Fynbo et al. (2001) find that ^ 3/4 of the limiting magnitudes for the dark bursts are brighter 
than the detected, but faint, optical afterglow of GRB 000630, and conclude that the dark bursts 
are consistent with a failure to image deeply enough quickly enough. However, Lazzati, Covino 
& Ghisellini (2001), using a smaller sample of dark bursts, find that the distribution of limiting 
magnitudes for the dark bursts, even if treated as detections, is significantly fainter than the 
distribution of magnitudes of the detected optical afterglows, and conclude that the dark bursts 
cannot be explained by a failure to image deeply enough quickly enough. Lazzati, Covino & 
Ghisellini (2001) also find that the X-ray fluxes of the dark bursts are not significantly different 
than the X-ray fluxes of the detected optical afterglows. However, we find the opposite below. 

We resolve these issues by applying Bayesian inference, a statistical formalism in which limits 
can be treated as limits, instead of detections (e.g., Reichart 2001a; Reichart et al. 2001). We 
resolve the issue of whether dark bursts can be explained by a failure to image deeply enough 
quickly enough in §2. Applying the same statistical formalism in §§3 and 4, we show that dark 
bursts tend to be X-ray and radio faint. In §5, we show that these correlations, and more 
specifically the optical vs. X-ray and optical vs. radio distributions, can be explained within the 
framework of the relativistic fireball model if the dark bursts are extinguished by circumbursly 
dust, and the density of the circumburst medium spans many orders of magnitude. We address a 
number of alternative explanations in §6. We draw conclusions in §7. 



2. Evidence that the Majority of Bursts are Dark 

The samples of Fynbo et al. (2001) and Lazzati, Covino & Ghisellini (2001) differ in the 
following ways: The sample of Fynbo et al. (2001) is nearly complete - it contains limiting 
magnitudes for 95% of the dark bursts preceding GRB 000630 - whereas the sample of Lazzati, 
Covino & Ghisellini (2001) includes limiting magnitudes for BeppoSAX bursts only. Also, 
the sample of Lazzati, Covino & Ghisellini (2001) relies less on GCN Reports, and more on 
published results. As BeppoSAX bursts appear to have been deeply imaged more often than 
bursts detected by other satellites, the different findings of these two papers can be attributed to 
sample differences, and the practice of comparing limits to detections, which can turn such sample 
differences into sample biases, as we demonstrate below. 



^We define 'dark burst' to mean a rapidly-, well-localized burst with an undetected optical afterglow. 

By 'circumburst', we mean within the circumburst cloud, which is probably parsecs to tens of parsecs across 
(Reichart & Price 2001; see also Galama & Wijers 2001; Reichart 2001b). 



But first, we combine the samples of Fynbo et al. (2001) and Lazzati, Covino &: Gliisellini 
(2001), keeping only the bursts through GRB 000630 to maintain the completeness of the Fynbo 
et al. (2001) sample. As is done in these papers, we scale the data to a common time: We use 
a temporal index of -1.4, the median temporal index in Table 1 of Lazzati, Covino & Ghisellini 
(2001), and we scale the data to 18 hours after the burst, the median observation time in our 
combined sample. In the event of multiple limiting magnitudes for a single burst, we adopt 
the most constraining. We plot the combined sample in Figure 1: The hashed histogram is a 
binning of the R-band magnitude distribution of the detected optical afterglows. The unhashed 
histogram, which is added to the hashed histogram, is a binning of the limiting R-band magnitude 
distribution of the dark bursts. 

Although these distributions appear to be similar, since one is a distribution of limits, they 
can be consistent only if the vast majority of the dark bursts could have been detected had 
they only been imaged a little more deeply. Being this unlucky, consistently, is of course very 
improbable. We quantify this improbability with Bayesian inference: Let n(R) be the normalized 
R-band magnitude distribution of all of the bursts. The likelihood function is then given by 



N 



where A^ is the number of bursts in our combined sample, and 

Ci-- 



(1) 



n(Rj) (Rj = detection) 

J-^ n(R)dR (Ri = hmit) 



(2) 



(e.g., Reichart 2001a; Reichart et al. 2001). We now consider a wide variety of two-parameter 
models for n(R): a Gaussian, 



n(R) 



1 



27r6 



exp 



R 



a boxcar, 





1 

b—a 





an increasing power law. 



and a decreasing power law. 



-(R) = bh 



n(R) 



(R<a) 
(a < R < 6) 
(R>6) 




(R<0) 
{0<R<b) , 
(R>6) 



(3) 



(4) 



(5) 



n(R) 



{R<b) 
-f^R" {R>b) ■ 



(6) 



We have fitted these models to the data in Figure 1, we plot the best-fit models in Figure 1 
(dotted curves), and we list the best-fit parameter values and 68% credible intervals in Table 1. 



Also in Table 1, we list the fractions /all of all bursts, and /dark of dark bursts, that have 
afterglows fainter than R = 24 mag 18 hours after the burst, the magnitude scaled to this time of 
the faintest detected optical afterglow in our sample. The results are fairly independent of the 
assumed shape of the brightness distribution: On average, we find that ~ 57_^]^% of all bursts, 
and ~ 82l;^7% of dark bursts, have afterglows that are fainter than R = 24 mag 18 hours after 
the burst. Consequently, the dark bursts cannot be explained by a failure to image deeply enough 
quickly enough: As a whole, they are fainter than the detected optical afterglows. 



3. Evidence that Dark Bursts Tend to be X-ray Faint 

Lazzati, Covino & Ghisellini (2001) argue that while the optical afterglows of the dark 
bursts are significantly fainter than the detected optical afterglows, their X-ray afterglows are not 
similarly faint. We adopt the Lazzati, Covino & Ghisellini (2001) sample of 2 - 10 keV afterglow 
fluxes, through GRB 000630 (§2), and add to it the 2-10 keV afterglow fluxes of non-BeppoSAX 
dark bursts from a review of lAU Circulars and GCN Reports: In particular, we have added X-ray 
data for the RXTE bursts GRB 970616 (Marshall et al. 1997), GRB 970815 (Murakami et al. 
1997a), and GRB 970828 (Murakami et al. 1997b), where the latter two were observed by ASCA. 
As we did in §2, we scale the data to a common time: We again use a temporal index of -1.4, and 
we scale the data to 9 hours after the burst, the median observation time in our expanded sample. 
We plot these data vs. the optical data from our combined sample (§2) in Figure 2. We notice 
that most of the dark bursts have 2-10 keV afterglow fluxes between w 10^^^ and ~ 10^^^ erg 
cm~^ s~^, while most of the bursts with detected optical afterglows have 2-10 keV afterglow 
fluxes between ~ 10^^^ and ~ 10^^^ erg cm~^ s~^. 

We use Bayesian inference to estimate the probability that the optical and X-ray data are 
correlated. Let n(R, logF) be the normalized distribution of the data in the R - log-F plane. The 
likelihood function is again given by Equation ([l|), but with Ci now given by 

' n(Ri,logFj) (Rj = detection, Fi = detection) 

/^ ?7-(R, log-Fj)(iR (Rj = limit, Fi = detection) 

/i°^^' n(Ri , log F)dlog F (Ri = detection, Fi = limit) ^'^^ 

, /~ /i°^^' n(R, log F)dRd log F (R^ = limit, Fi = limit) 



Cr 



(e.g., Reichart 2001a; Reichart et al. 2001). In this section, we consider only a single, simple 
model for n(R, log-F): a two-dimensional Gaussian rotated an angle 9 in the R - logF plane; i.e., 
contours of constant n(R, logF) are simply ellipses rotated in the R - log-F plane. We consider a 
physically motivated model in §5. We plot the best-fit model (solid ellipse) in Figure 2. Clearly, a 
positive correlation (0 < ^ < '7r/2) is favored. One measure of how significantly < 9 < tt/2 is the 
probability that —it/2 < 9 < 0. We find that negative correlations are ruled out at the ~ 2.7 a 
confidence level. This becomes more significant when we consider the physically motivated model. 

One might argue that GRB 970828, the poster child of dark bursts (Groot et al. 1998; 



Djorgovski et al. 2001), dominates the fit. Consequently, we redo the fit excluding GRB 970828, 
and the correlation remains (dotted ellipse). Clearly, the numerous dark bursts with faint X-ray 
afterglows dominate the fit. 



4. Evidence that Dark Bursts Tend to be Radio Faint 

We also address the question of whether the radio afterglows of the dark bursts are similarly 
faint. We have reviewed the literature, and present our sample of radio flux densities and limits 
for 25 bursts in Table 2. We constructed this sample in the following way: If available, we 
included the ~ 8.46 GHz datum closest to 10 days after the burst. We selected 8.46 GHz to 
reduce contamination by interstellar scintillations (e.g., Goodman 1997; Frail et al. 1997), and 
because it is a commonly used frequency. We selected 10 days to reduce contamination by reverse 
shocks (e.g., Sari & Piran 1999; Kulkarni et al. 1999), and again to reduce contamination by 
interstellar scintillations. Ten days is also good because it typically proceeds tjet (Frail et al. 
2001), and consequently the light curve is either constant or slowly fading until the synchrotron 
frequency i^m < 8.46 GHz (e.g., Sari, Piran &: Halpern 1999; Harrison et al. 1999): Hence, our 
results are fairly independent of the choice of 10 days. However, many of the undetected radio 
afterglows did not have limits available at this frequency or time. If frequencies were available 
only at ~ 4.86 GHz, we scaled the limits to 8.46 GHz using a spectral index of +2, the expected 
spectral index if the self-absorption frequency z/q > 8.46 GHz (e.g., Sari, Piran & Narayan 1998): 
A spectral index of -1-1/3 is expected if i^a < 8.46 GHz (e.g.. Sari, Piran & Narayan 1998), so we 
have adopted the more conservative scaling index. If times were available only preceding 10 days, 
we scaled the latest limit to 10 days using a temporal index of -|-1, the expected temporal index if 
the circumburst medium is the wind of a massive star and Va > 8.46 GHz (Chevalier & Li 2000): 
A temporal index of zero is expected if the circumburst medium is the wind of a massive star and 
i^a < 8.46 GHz (Chevalier & Li 2000), and a temporal index of -1-1/2 is expected if the circumburst 
medium is constant in density, independent of whether i^a > 8.46 GHz or i^a < 8.46 GHz (e.g., 
Sari, Piran & Narayan 1998), so again we have adopted the conservative scaling index. Finally, all 
of the limits in Table 2 have been converted to ~ 2 o" before scaling. 

For two bursts, GRB 970828 (Figure 3) and GRB 990123 (Figure 4), we can do better. 
Although no observation after a few days after the burst resulted in a definitive detection, nearly 
all of the measurements out to ~ 100 dy after the burst favor positive flux densities, suggesting 
that the afterglow is present at late times, but at a faint level. Kulkarni et al. (1999) found 
this to be the case for GRB 990123 by binning the late-time light curve. Here we take another 
approach: We simultaneously fit the forward shock model of Sari, Piran & Halpern (1999) and the 
reverse shock model of Sari & Piran (1999) to the data, and, allowing for interstellar scintillations, 
constrain the flux density of the forward shock 10 days after the burst. In the case of GRB 970828, 
we find that F8.46GHz(At = 10 dy) = 16lg iJ,Jy, with a median value of 27 ;uJy, and we detect 
the forward shock at the ~ 3.0 a level confidence level. In the case of GRB 990123, we find that 



-p8.46GHz(At = 10 dy) = 161q ^Jy, with a median value of 18 fiJy, and we detect the forward shock 
at the ~ 3.2 a level confidence level. We use the median values in the following fits. 

We plot our radio sample vs. our combined optical sample (§2) in Figure 5. We notice 
that most of the optically dark bursts are also radio dark. However, we leave it to the analysis 
that we presented in §3 to decide whether these data are correlated: Indeed, we again find that 
a positive correlation is favored (solid ellipse), and we find that negative correlations are ruled 
out at the ~ 2.6 a confidence level. Again, this becomes more significant when we consider a 
physically motivated model in §5. Also, we again show that exclusion of GRB 970828 does not 
significantly affect the result (dotted ellipse). Clearly, the numerous dark bursts with undetected 
radio afterglows dominate the fit. 



5. Evidence that Dark Bursts are Extinguished by Circumburst Dust and that the 
Density of the Circumburst Medium Spans Many Orders of Magnitude 

If the dark bursts were the result of extinction by dust unrelated to the circumburst medium, 
e.g., dust elsewhere in the host galaxy, one would expect very little correlation in Figures 2 and 
5, since the optical properties of the afterglow would be dominated by this unrelated dust, and 
the X-ray and radio properties of the afterglow would be determined by the physical properties of 
the relativistic fireball and circumburst medium. However, if the dark bursts were the result of 
extinction by circumburst dust, correlations might be expected, since the optical, X-ray, and radio 
properties of the afterglow would all depend on the density of the circumburst medium (e.g.. Sari, 
Piran & Narayan 1998; Sari & Esin 2001; Harrison et al. 2001). Consequently, in this section, we 
determine whether the observed correlations, and more specifically the observed optical vs. X-ray 
and optical vs. radio distributions, can be reproduced theoretically, using the afterglow model of 
Sari & Esin (2001) to which we add extinction by circumburst dust (e.g., Reichart 2001a), by 
varying the density of the circumburst medium while fixing the remaining parameters to canonical 
values. 

However, this is complicated by the fact that the optical flash (e.g.. Sari & Piran 1999; 
Akerlof et al. 1999) is expected to destroy circumburst dust via sublimation, and the burst and 
afterglow are expected to change the size distribution of the circumburst dust via fragmentation 
(e.g., Waxman Sz Draine 2000; Galama & Wijers 2001; Fruchter, Krolik & Rhoads 2001; Reichart 
2001b). This is further complicated by the fact that the distances to which the optical flash 
sublimates dust, and the burst and afterglow fragment dust, are functions of grain size: The 
sublimation distance is a decreasing function of grain size, while the fragmentation distance is 
a decreasing function of grain size for large grains and an increasing function of grain size for 
small grains (e.g., Waxman & Draine 2000; Fruchter, Krolik & Rhoads 2001; Reichart 2001b). In 
Reichart (2001b), we show that for a canonical distribution of graphite and silicate grain sizes, and 

1/2 

a simple fragmentation model, the burst, optical flash, and afterglow burn through ~ 10^49 P*^ °^ 
optical depth, where L49 is the 1 - 7.5 eV isotropic-equivalent peak luminosity of the optical flash 



in units of 10^^ erg s~^ (the 1 - 7.5 eV isotropic-equivalent peak luminosity of the optical flash of 
GRB 990123; Waxnian & Draine 2000). Furthermore, this distance is fairly independent of the 
density of the circumburst cloud, and of the isotropic-equivalent energy of the burst and afterglow. 

We now consider under what conditions bursts burn completely through their circumburst 
clouds. Frail et al. (2001) show that the bursts for which redshifts have been measured draw 
upon a fairly standard energy reservoir of ~ 3 x 10^^ erg (for an efficiency at which this energy 
is converted to gamma rays of 77 ~ 0.2; e.g., Beloborodov 2000), and that the wide range of 
isotropic-equivalent energies that have been implied for these bursts, from ^ 3 x 10^^ erg to 
;^ 3 X 10^^ erg, is primarily the result of a wide range of collimation angles, with half angles 
ranging from <^ 0.05 rad to ^ 0.5 rad. If this is indeed the case, one expects a wide range of 
isotropic-equivalent peak luminosities for the optical flashes of these bursts, and consequently 
a wide range of optical depth burn distances, ranging from parsecs to several tens of parsecs. 
Consequently, since most clouds, including most giant molecular clouds, tend to be less than 
several tens of parsecs across (e.g., Solomon et al. 1987), strongly collimated bursts likely burn 
completely through their circumburst clouds, while weakly collimated bursts likely often do not]^ 

Consequently, we consider two cases: a strongly collimated burst, and a weakly collimated 
burst. In the case of the strongly collimated burst, we assume that the burst, optical flash, and 
afterglow burn completely through the optical depth of the circumburst cloud. In the case of the 
weakly collimated burst, we assume that a few parsecs of dust remain. 



Specifically, we adopt the following parameter values: Using the now-common notation of 
Sari, Piran & Narayan (1998) and Sari, Piran &; Halpern (1999), in the case of the strongly 

9j 



collimated burst, we adopt E = 3 x 10^^ erg and 9jet = 0.05 rad, and in the case of the weakly 



collimated burst, we adopt E = 3 x 10^^ erg and Oj^t = 0.5 rad. For the remaining relativistic 
fireball parameter values, we adopt eg = 0.3, €b = 0.01, and p = 2.4, based on the best fit of the 
model of Sari & Esin (2001) to the afterglow of GRB 000926 (Harrison et al. 2001). We have 
chosen this fit of this model to this afterglow, because it is the first such fit to include (and detect) 
the inverse-Compton component of the afterglow (see below). We adopt a redshift of z = 1, 
because it is typical of the redshifts that have been measured for the bursts to date. Finally, we 
vary the density of the circumburst cloud from n = 10~^ cm~'^, which is typical of the Galactic 
disk, to n = 10^ cm~^, which is typical of dense clouds, and in the case of the weakly collimated 
burst, we vary the number of parsecs of circumburst dust that remain from 1 to 10. 

Using the same code that Harrison et al. (2001) use, we plot the optical vs. X-ray results in 
Figure 6, and the optical vs. radio results in Figure 7. Qualitatively, these curves can be explained 



^If the circumburst cloud is the nuclear region of an ultraluminous infrared galaxy, as has been proposed by 
Ramirez-Ruiz, Trentham & Blain (2001) to explain the dark bursts, even strongly collimated bursts would not burn 
through the hundreds of parsecs of optical depth that are typical of such regions (e.g., Solomon et al. 1997) . However, 
we show in Reichart & Price (2001) that the limited information that is available on the column densities, measured 
from absorption of the X-ray afterglow, of the dark bursts in not consistent with this idea. 



as follows: At a circumburst cloud density of n = 10^^ cm^^, the ordering of the self-absorption 
frequency f^, synchrotron frequency u^, and cooling frequency Uc is z/q < 8.46 GHz < Vm 10 
days after the burst, fm < ^R < ^c 18 hours after the burst, and z^^ < 2 - 10 keV 9 hours after 
the burst. As we increase the density of the circumburst cloud to n = 10 — 10^ cm~^, the radio 
afterglow grows brighter, because the brightness of the spectral peak {vm at these densities), and 
consequently the brightness of the afterglow between Va and fc, increases with increasing density 
(e.g.. Sari, Piran & Narayan 1998; Sari & Esin 2001). The optical afterglow also grows brighter 
for this reason, but only until n = 1 — 10 cm"'^. At higher densities, i^c, which decreases with 
increasing density (e.g.. Sari, Piran & Narayan 1998; Sari & Esin 2001), passes through the optical, 
resulting in a net decrease in brightness. Since z^c < 2 - 10 keV for all of these densities, the X-ray 
afterglow decreases in brightness for the same reason, but only until n ~ 10 cm~^. At higher 
densities, the inverse-Compton component, which grows brighter with increasing density (e.g.. 
Sari & Esin 2001), dominates. This is counteracted once again at very high densities (the exact 
value depends on the adopted value of E), because i^a, which increases with increasing density 
(e.g.. Sari, Piran & Narayan 1998; Sari & Esin 2001), passes z/^ and Vc, causing the brightness of 
the spectral peak (f^ at these densities), and consequently the brightness of the afterglow at all 
frequencies, to decrease with increasing density. 

As we increase the density of the circumburst cloud beyond n = 10 — 10^ cm~'^, the radio 
afterglow decreases in brightness, because i^a passes through the radio. Finally, in the case of 
the weakly collimated afterglow, as we increase the density beyond n ~ 10^ cm~^, the optical 
afterglow decreases in brightness substantially, since extinction scales as R ~ n, where all of the 
above effects scale only logarithmically with n. 

We now investigate whether such a two-component model is strongly requested by the data. 
First, we investigate whether a general, two-component model is strongly requested by the data, 
and then we check whether the best-fit model is in at least broad, qualitative agreement with the 
theoretical expectation. Again, we apply Bayesian inference. The likelihood function is again given 
by Equations (|l]) and (|^), but we now adopt a simple two-component model for n(R, logF): the 
sum of two two-dimensional Gaussians, each rotated in the R - logF plane, where one Gaussian 
has weight w, and the other has weight 1 — w. We plot the best-fit models (solid ellipses) in 
Figures 8 and 9. In the case of the optical vs. X-ray distribution, the two-component model is 
favored over the one-component model of §3 at the « 3.7 a confidence level (despite the additional 
model parameters), and it is favored over the correlation-free model of §3 at the ~ 4.1 o" confidence 
level. In the case of the optical vs. radio distribution, the two-component model is favored 
over the one-component model of §4 at the ~ 3.1 o" confidence level, and it is favored over the 
correlation-free model of §4 at the ~ 3.5 a confidence level. Consequently, two-component models 
are strongly requested by these data, and one-component and correlation-free models are strongly 
ruled out. Again, we show that exclusion of GRB 970828 does not significantly affect these results 
(dotted ellipses). 

Finally, we check whether the best-fit models of Figures 8 and 9 are in at least broad. 



qualitative agreement with the theoretical expectations of Figures 6 and 7. To this end, we 
have replotted the best-fit models in Figures 6 and 7, and, frankly, the agreement is remarkable. 
Figures 6-9 suggest that if the afterglows of these bursts are similar in their other properties to 
the afterglow of GRB 000926, the densities of their circumburst clouds span orders of magnitude, 
from densities that are typical of the Galactic disk to densities that are typical of dense clouds. 
In particular, Figures 6-9 suggest that if the afterglows of GRB 970828 and GRB 990123 (R 
Ki 20.6 mag, i^2-iokcV ~ 5.9 x 10^^^ erg cm~^ s~^, F8.46gHz ~ 18 /xJy) are similar in their other 
properties to the afterglow of GRB 000926, the densities of their circumburst clouds are of the 
order n ~ 10^ — 10^ cm~^. In the case of GRB 990123, this is a prediction that modeling of the 
afterglow data should be able to either confirm or refute]^ 



6. Evidence Against Alternative Explanations 

We now address a number of alternative explanations: 

Galactic extinction: Lazzati, Covino & Ghisellini (2001) show that the distribution of 
Galactic column densities for the dark bursts is consistent with the distribution of Galactic column 
densities for the bursts with detected optical afterglows, ruling this out as an explanation for the 
vast majority of the dark bursts. 

Host galaxy extinction unrelated to the circumburst medium: As we discuss in §5, 
if this were the case, one would expect very little correlation in Figures 2, 5, 8, and 9. However, 
this does not appear to be the case (§§3, 4, and 5), ruling this out as an explanation for perhaps 
all but a few of the dark bursts. 

High redshift eflfects: Lyman limit absorption in the source frame, absorption by the Lya 
forest, absorption by excited molecular hydrogen in the circumburst medium (Draine 2000), and 
source- frame extinction by the FUV component of the extinction curve (e.g., Reichart 2001a) could 
all result in dark bursts if at sufficiently high redshifts. However, Lamb & Reichart (2000) show 
that unless the burst history of the universe differs dramatically from the star-formation history of 
the universe, the redshift distribution of the bursts should be fairly narrowly peaked around z ~ 2, 
primarily because there is very little volume in the universe at low and high redshifts. Weinberg 
et al. (2001) model the detection efficiency functions of BeppoSAX and IPN, and show that these 
satellites should detect even fewer bursts at high redshifts, pushing the expected typical redshift 
for bursts detected by these satellites down to the observed value of about one. However, based on 
their variability redshift estimates of 220 BATSE bursts, Fenimore &: Ramirez-Ruiz (2001; see also 
Schaefer, Deng &; Band 2001) find that the burst history of the universe might differ dramatically 



^Panaitescu & Kumar (2001) model the afterglow data of GRB 990123, and find that n ~ 5 x 10^* cm^^. However, 
the data might be better accommodated by a cooling break between the optical and X rays, and a sufficiently dense 
circumburst medium to introduce an inverse-Compton component to accommodate the X-ray data. 



10 



from the star-formation history of the universe, with very many more bursts at high redshifts. 
However, using similar variabihty redshift estimates (Reichart et al. 2001) for 907 BATSE bursts, 
Reichart & Lamb (2001) find that only ~ 15% of bursts above BATSE's detection threshold 
have z > 5 (and if the luminosity function is evolving, far fewer bursts below BATSE's detection 
threshold have z > 5). Consequently, the above high redshift effects probably affect ^ 10% of the 
bursts in our BeppoSAX- and IPN-dominated sample. Furthermore, Ramirez-Ruiz, Trentham & 
Blain (2001) find that the variability redshift estimates for all of the dark bursts for which high 
resolution BATSE light curves are available have z < 5. 

A second class of long-duration bursts vi^ith afterglows that are described by a 
very different parameterization of the relativistic fireball model: The postulation of a 
second class of objects to explain a phenomenon that can be explained more simply is always ad 
hoc. However, for the sake of argument, we investigate whether the relativistic fireball model 
can account for a dark burst like GRB 970828 without calling upon extinction, and if so, what 
properties the relativistic fireball and circumburst medium must have. To this end, we have fitted 
the model of Sari & Esin (2001) to the following data: R > 23.8 mag 0.14 days after the burst 
(Groot et al. 1998), -F2-iokcV = 4 x 10"-*^^ erg cm~^ s~^ 1.2 days after the burst (Murakami et 
al. 1997b), and -F8.46GHz = 16^g fiJy 10 days after the burst (§4). Given that the number of 
degrees of freedom is negative, one would expect a x^ of zero, and a broadly degenerate solution. 
Instead, we find a x^ of a few, and the following solution: £■ ~ 3 x 10^^ erg, eg ~ 0.3, es ~ 10~^, 
n ~ 3 X 10^ cm~'^, and 6 jet ~ 7r/2 (isotropic). These parameter values can be explained as 
follows: The high value of n increases the brightness of the inverse-Compton component of the 
afterglow to accommodate the high X-ray to optical and X-ray to radio relative brightnesses of the 
afterglow. However, unless compensated for elsewhere, the high value of n would also (1) push the 
self-absorption frequency significantly above 8.46 GHz, making the radio afterglow undetectable, 
and (2) push the jet break time significantly before 1.2 days after the burst, making the X-ray 
afterglow too faint. The model compensates for (1) with an extremely low value of €b, and a value 
of E that is about an order of magnitude greater than the expectation from Frail et al. (2001), 
given that the fireball is isotropic. The model compensates for (2) by making the fireball isotropic. 
In any case, although the fitted model cannot be ruled out on physical grounds, it does not fit the 
data as well as might be expected, and is unavoidably ad hoc. We note that we have also tried the 
wind model of Chevalier & Li (2000) with similar results. 



A second class of long-duration bursts with afterglows that do not arise from 
relativistic fireballs: Once again, the postulation of a second class of objects to explain a 
phenomenon that can be explained more simply is ad hoc. However, this postulation is particularly 
ad hoc: Whereas in the previous postulation, we abandoned only a canonical parameterization 
of a successful physical model for a very different parameterization of the same model, in this 
postulation, we are abandoning all parameterizations of a successful physical model for an 
unknown physical model. We include this possibility only for the sake of completeness. 



11 



7. Discussion and Conclusions 

We have shown that about 60% of all bursts, and about 80% of dark bursts, have afterglows 
that are fainter than R = 24 mag 18 hours after the burst. Furthermore, we have shown that, with 
the exception of perhaps a few bursts, the dark bursts are most likely the result of circumburst 
extinction, and the density of the circumburst medium probably spans many orders of magnitude, 
from densities possibly as low as densities that are typical of the Galactic disk to densities probably 
as high as densities that are typical of dense clouds. 

Frail et al. (2001) we show that the distribution of values of En~^''^7f'''^ spans a range of 
a factor of only Q.b^i^ (90% width), where t] is the efficiency at which the total energy E is 
converted to gamma rays. Consequently, the density n of the circumburst medium is constrained 
to a range of a factor of ^ 10 ' -o-". This is perhaps an order of magnitude or two smaller than 
the range of densities that we find in this paper, but consistent given the uncertainties. However, 
there is a selection effect that brings these two estimates into better agreement, and that is that 
the distribution of values of En~^'^r]^'^ is necessarily constructed using only bursts for which the 
redshift has been measured, and consequently dark bursts are selected against: Dark bursts for 
which the redshift has been measured from emission-line spectroscopy of the host galaxy via a 
radio localization (e.g., GRB 970828; Djorgovski et al. 2001) are rare, probably because optically 
dark bursts tend to also be radio dark (§4). Since dark bursts probably have n ^ 10^ cm~^ (§5), 
the true distribution of values of En^^''^ri'^''^ is probably somewhat broader. 

Furthermore, since the dark bursts are probably more weakly collimated than the bursts with 
detected optical afterglows (§5), the true distribution of opening angles is probably somewhat 

shallower than 0- ^ "■®, the distribution that Frail et al. (2001) find. Because of this, the true 
burst rate is probably somewhat greater than 520 it 85 times the observed rate, and consequently 
neutron star coalescence becomes even more strongly disfavored than what Frail et al. (2001) find. 

Support for this work was provided by NASA through the Hubble Fellowship grant 
#HST-SF-01133.01-A from the Space Telescope Science Institute, which is operated by the 
Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555. 
We are also grateful to Dale Frail, Johan Fynbo, Don Lamb, Alin Panaitescu, and Paul Price for 
useful discussions and information. 



12 



REFERENCES 

Akerlof, C, et al. 1999, Nature, 398, 400 

Beloborodov, A. M., 2000, ApJ, 539, 25 

Berger, E., et al. 2000, ApJ, 545, 56 

Berger, E., et al. 2001, ApJ, submitted (|astro-ph/0102278D 

Berger, E., & Frail D. A. 2000a, GCN Report 686 

Berger, E., & Frail D. A. 2000b, GCN Report 738 

Chevalier, R. A., & Li, Z. 1999, ApJ, 520, L29 

Djorgovski, S. C, et al. 2001, ApJ, submitted (astro-ph) 

Draine, B. T. 2000, ApJ, 532, 273 

Fenimore, E. E., k Ramirez-Ruiz, E. 2001, ApJ, submitted ( ^stro-ph/0004176| ) 

Frail, D. A. 1999, GCN Report 195 



Frail 
Frail 
Frail 
Frail 
Frail 
Frail 
Frail 
Frail 
Frail 
Frail 
Frail 
Frail 
Frail 



D. A. 2000a, GCN Report 619 
D. A. 2000b, GCN Report 622 
D. A., Becker, K. M., & Berger, E. 2000, GCN Report 721 



D. A. 
D. A. 
D. A. 
D. A. 
D. A. 
D. A. 
D. A. 
D. A. 
D. A. 
D. A. 



et al. 1997, Nature, 389, 261 

et al. 1998a, GCN Report 141 

et al. 1998b, ApJ, 502, L119 

et al. 2000a, ApJ, 534, 559 

et al. 2000b, ApJ, 538, L129 

et al. 2001, Nature, submitted (|astro-ph/0102282|) 

& Kulkarni, S. R. 1997, GCN Report 7 

k Kulkarni, S. R. 1998, GCN Report 170 

k Taylor, G. B. 2000, GCN Report 574 

Waxman, E., k Kulkarni, S. R. 2000, ApJ, 537, 191 



Fruchter, A. S., Krolik, J. H., k Rhoads, J. E. 2001, ApJ, in press (|astro-pli/0106343|) 



- 13- 

Fynbo, J. U., et al. 2001, A&A, 369, 373 

Galama, T. J., et al. 2000, ApJ, 541, L45 

Galama, T. J., k Wijers, R. A. M. J. 2001, ApJ, 549, L209 

Groot, P. J., et al. 1998, ApJ, 493, L27 

Goodman, J. 1997, New Astronomy, 2, 449 

Harrison, F. A., et al. 1999, ApJ, 523, L121 

Harrison, F. A., et al. 2001, ApJ, in press ( Mtro-ph/0103377l ) 

Kulkarni, S. R., et al. 1999, ApJ, 522, L97 

Lamb, D. Q., & Reichart, D. E. 2000, ApJ, 536, 1 

Lazzati, D., Covino, S., & Ghisellini, G. 2001, MNRAS, submitted ( |astro-ph/0011443| ) 

Marshall, F. E., et al. 1997, lAU Circular 6683 

McConnell, D., et al. 2000, GCN Report 560 

Murakami, T., et al. 1997a, lAU Circular 6722 

Murakami, T., et al. 1997b, lAU Circular 6732 

Panaitescu, A., & Kumar, P. 2001, ApJ, in press ( |astro-ph/0010257| ) 

Ramirez-Ruiz, E., Trentham, N., & Blain, A. W. 2000, MNRAS, submitted (|astro-ph/0103239| ) 

Reichart, D. E. 2001a, ApJ, 553, 235 

Reichart, D. E. 2001b, ApJ, submitted (astro-ph) 

Reichart, D. E., et al. 2001, ApJ, 552, 57 

Reichart, D. E., & Lamb, D. Q. 2001, in Procs. of the 20th Texas Symposium on Relativistic 



Astrophysics, in press (|astro-ph/0103255D 



Reichart, D. E., & Price, P. A. 2001, ApJ, submitted (astro-ph) 

Rol, E., et al. 1999, GCN Report 374 

Sari, R., & Esin, A. A. 2001, ApJ, 548, 787 

Sari, R., & Piran, T. 1999, ApJ, 520, 641 

Sari, R., Piran, T., & Halpern, J. P. 1999, ApJ, 519, L17 



- 14- 

Sari, R., Piran, T., & Narayan, R. 1998, ApJ, 497, L17 

Schaefer, D. E., Deng, M., & Band, D. L. 2001, ApJ, submitted ( |astro-ph/0101461| ) 

Solomon, P. M., et al. 1987, ApJ, 319, 730 

Solomon, P. M., et al. 1997, ApJ, 478, 144 

Taylor, G. B., et al. 1998, ApJ, 502, L115 

Taylor, G. B., et al. 2000, ApJ, 537, L17 

Taylor, G. B., Frail, D. A., & Kulkarni, S. R. 1999, GCN Report 425 

Wark, R., et al. 1999, GCN Report 266 

Waxman, E., & Draine, B. T. 2000, ApJ, 537, 796 

Weinberg, N., et al. 2001, in preparation 



This preprint was prepared with the AAS lATp^X macros v4.0. 



15 



Table 1. R-Band Magnitude Distribution Model Fits and Faint Burst Fractions 



Function a b /all /dark 

Gaussian 23.98to-57 2.951^^ 0.50l[j-^^ 0-7'^^oil 

Boxcar 17.12t°:!]3 34.40lll° 0.60t°i^ o.86t°J? 

Increasing Power Law 5.74^1^1 26.83tlil 0.53+13:^^ O-'^^-olt 

Decreasing Power Law -2.23toil 17.14+g:^^ OM^oii OM^oij 

Average - - 0.57t[j:i? 0.82to;?7 



16 



Table 2. Radio Data 



GRB 


u [GHz] 


At [dy] 


Fu [^Jy] 


pscaled [^jy] 


Ref(s)^ 


970228 


8.46 


11.0 


<80 


<80 


1 


970508 


8.46 


9.95 


610 ±51 


610 


2 


970828 


8.46 


10 


16tf 


27 


3,4 


971214 


8.46 


1.37 


<34 


<248 


5 


980329 


8.46 


9.90 


179 ± 41 


179 


6 


980519 


8.46 


14.05 


142 ± 29 


142 


7 


980703 


8.46 


5.32 


965 


965 


8 


981220 


8.46 


6.30 


<72 


< 114 


9 


981226 


8.46 


8.54 


169 ± 28 


169 


10 


990123 


8.46 


10 


161^ 


18 


3, 11 


990217 


8.6 


1.09 


< 112 


< 1028 


12 


990506 


8.4 


15.66 


< 137 


< 137 


13 


990510 


8.6 


9.22 


127 ±31 


127 


14 


990704 


4.88 


1.02 


<65 


< 1915 


15 


991014 


8.46 


1.68 


<50 


<298 


16 


991208 


8.46 


9.62 


999 ± 50 


999 


17 


991216 


8.46 


9.73 


170 ± 72 


170 


18 


000210 


8.46 


4.60 


<55 


< 120 


19 


000301A 


8.46 


2.75 


< 136 


<495 


20 


000301C 


8.46 


12.17 


483 ± 26 


483 


21 


000326 


4.86 


1.48 


<41 


<839 


22, 23 


000418 


8.46 


9.48 


100 ± 25 


100 


24 


000528 


8.46 


3.74 


<80 


< 214 


25 


000615 


8.46 


4.87 


<75 


< 154 


26 


000630 


4.86 


1.02 


< 100 


<2971 


27 



*1. Frail et al. 1998b; 2. Frail, Waxman & Kulkarni 2000; 3. this 
paper; 4. Djorgovski et al. 2001; 5. Frail & Kulkarni 1997; 6. Taylor 
et al. 1998; 7. Frail et al. 2000a; 8. Frail et al. 1998a; 9. Frail & 
Kulkarni 1998; 10. Frail 1999; 11. Kulkarni et al. 1999; 12. Wark et 
al. 1999; 13. Taylor et al. 2000; 14. Harrison et al. 1999; 15. Rol 
et al. 1999; 16. Taylor, Frail & Kulkarni 1999; 17. Galama et al. 
2000; 18. Frail et al. 2000b; 19. McConnell et al. 2000; 20. Frail & 
Taylor 2000; 21. Berger et al. 2000; 22. Frail 2000a; 23. Frail 2000b; 
24. Berger et al. 2001; 25. Berger k Frail 2000a; 26. Frail, Becker & 
Berger 2000; 27. Berger & Frail 2000b. 



17 



Fig. 1. — R-band magnitude distribution of the detected optical afterglows (hashed histogram), 
and limiting magnitude distribution of the dark bursts (unhashed histogram, added to the hashed 
histogram), scaled to 18 hours after the burst. The dotted curves are the best-fit models of §2 to 
these data. Clearly, most bursts have afterglows fainter than R ~ 24 mag 18 hours after the burst, 
independent of the assumed shape of the brightness distribution. 

Fig. 2. — R-band magnitude or limiting magnitude, scaled to 18 hours after the burst, vs. 2-10 
keV flux or limiting flux, scaled to 9 hours after the burst. The solid ellipse is from the best-flt 
model of §3 to all of the data, and the dotted ellipse is from the best-fit model to all of the data 
excluding GRB 970828. The ellipses should encompass ~ 90% of the data. The optical and X-ray 
data appear to be correlated, and clearly GRB 970828 does not dominate this correlation. 

Fig. 3.— 8.46 GHz fight curve of the afterglow of GRB 970828 from Djorgovski et al. (2001). The 
dashed curves are the best-fit reverse and forward shock components of the afterglow, and the solid 
curve is the sum of these components (§4). Although no observation after a few days after the 
burst resulted in a definitive detection, all of the measurements out to ~ 100 dy after the burst 
favor positive flux densities, suggesting that the afterglow is present at late times, but at a faint 
level. 

Fig. 4.— 8.46 GHz light curve of the afterglow of GRB 990123 from Kulkarni et al. (1999). The 
dashed curves are the best-fit reverse and forward shock components of the afterglow, and the solid 
curve is the sum of these components (§4). Although no observation after a few days after the 
burst resulted in a definitive detection, nearly all of the measurements out to ^ 100 dy after the 
burst favor positive flux densities, suggesting that the afterglow is present at late times, but at a 
faint level. 

Fig. 5. — R-band magnitude or limiting magnitude, scaled to 18 hours after the burst, vs. 8.46 
GHz flux density or limiting flux density, scaled to 10 days after the burst. The solid ellipse is 
from the best-fit model of §4 to all of the data, and the dotted ellipse is from the best-fit model 
to all of the data excluding GRB 970828. The ellipses should encompass « 90% of the data. The 
optical and radio data appear to be correlated, and clearly GRB 970828 does not dominate this 
correlation. 

Fig. 6. — Theoretical R-band magnitude vs. 2-10 keV flux as a function of circumburst cloud 
density (n = 10~^ — 10^ cm~^) for a strongly collimated burst that burns through the dust of its 
circumburst cloud (solid curve, top right), and for a weakly collimated burst that burns through all 
but one (top dotted curve, bottom left), three (solid curve, bottom left), and ten (bottom dotted 
curve, bottom left) parsecs of dust of its circumburst cloud (§5). The ellipses are replotted from 
Figure 8. 

Fig. 7. — Theoretical R-band magnitude vs. 8.46 GHz flux density as a function of circumburst 
cloud density (n = 10"^ — 10^ cm^^) for a strongly collimated burst that burns through the dust 



18 



of its circumburst cloud (solid curve, top), and for a weakly collimated burst that burns through 
all but one (top dotted curve, bottom), three (solid curve, bottom), and ten (bottom dotted curve, 
bottom) parsecs of dust of its circumburst cloud (§5). The ellipses are replotted from Figure 9. 

Fig. 8. — R-band magnitude or limiting magnitude, scaled to 18 hours after the burst, vs. 2-10 
keV flux or limiting flux, scaled to 9 hours after the burst. The solid ellipse is from the best-fit 
model of §5 to all of the data, and the dotted ellipse is from the best-fit model to all of the data 
excluding GRB 970828. The ellipses should encompass ^ 90% of the data. The optical and X-ray 
data appear to be in good, qualitative agreement with the theoretical expectation of Figure 6, and 
clearly GRB 970828 does not dominate the fit. 

Fig. 9. — R-band magnitude or limiting magnitude, scaled to 18 hours after the burst, vs. 8.46 
GHz flux density or limiting flux density, scaled to 10 days after the burst. The solid ellipse is 
from the best-fit model of §5 to all of the data, and the dotted ellipse is from the best-fit model 
to all of the data excluding GRB 970828. The ellipses should encompass ~ 90% of the data. The 
optical and radio data appear to be in good, qualitative agreement with the theoretical expectation 
of Figure 7, and clearly GRB 970828 does not dominate the fit. 



19 



m 



6 



8 



6 



4 - 



2 - 








36 



32 28 24 20 

R(t = 18 hr) [mag] 



16 



20 



18 - 



20 - 



CO 



CO 



v^ 



22 - 



24 - 



26 - 



28 - 



30 - 




10-14 10-13 10-12 10-11 10-10 

f^2-iokev(t = 9 hr) [erg cm-^ s-i] 



21 



200 



"1 — I — I — I I 1 1 1 1 1 — I — I — I I 1 1 1 



^ 100 



:l 



X 

o 

CO 
00 







GRB 970828 




I I I I I 1 1 1 1 I I I I I 1 1 1 



1 



10 
At [dy] 



100 



22 



300 - 






X 

o 



200 - 



^ 100 

CO 







1 — I — I I I 1 1 II 1 — I — I I I 1 1 II 1 — I — I II II I 



GRB 990123 




I I I I II II I I I I 1 1 1 II I I I I 1 1 II I 



0.1 



1 10 

At [dy] 



100 



23 



16 



18 



S^20h 



I I I I 1 1 1 1 — I — I I I I 1 1 1 1 — 1 — I I I I 1 1 1 1 — I — r~r 






22 



CO 24 - 
'' 26 



v^ 



28 



30 - 




<- 



V 



I I I I I I I I I l_L 



10 



100 



1000 



F8.46GHz(t = 10 dy) [/xJy] 



24 



16 



18 



CD 



22 - 



CO 24 



ci:; 



26 - 



28 



30 - 




10-14 10-13 10-12 10-11 10-10 

f^2-iokev(t = 9 hr) [erg cm-^ s-i] 



25 



16 


1 Mill 


1 1 1 1 1 


Ml 1 1 1 1 1 M 


1 1 




- 




10-' .-^ 


- 


18 






^^---^^^^y 


y^' 


t^20 

CO 




/--^ 


^^^^^^ 


:\ - 


6 

^22 


105^ 


^^^ — 


/^-^-^ 


1 


^ 24 




10-1 •r"'^ 


f/U 


~ 


CO 

^ ' 26 






/ J 10" J/ 




^28 




/ : 


/ . ./t/ 




Ci^ 


/ 


/ ; 


'' // ■■' 


- 


30 




■^ 


/ ^ 




32 


V 



1 1 :l 1 1 


III/ : 


1 1 



10 

F 



100 1000 

8.46GHz(t = 10 dy) [/xJy] 



26 



18 - 



20 - 



CO 



CO 



v^ 



22 - 



24 - 



26 - 



28 - 



30 - 




10-14 10-13 10-12 10-11 10-10 

f^2-iokev(t = 9 hr) [erg cm-^ s-i] 



27 



16 - 



18 



t^20 h 

CO 

^ 24 h 



CO 



V^ 



26 h 
28 
30 
32 h 



I I I I 1 1 1 1 — I — I I I I 1 1 1 1 — I — I I I I 1 1 1 1 — I — r~r 




<r 



Y 



I I I I I I I I I l_L 



10 



100 



1000 



F8.46GHz(t = 10 dy) [/xJy]